Open the model picker this morning and there's a gap where o3 used to sit. OpenAI retired it from ChatGPT on 26 August 2026, the end of a 90 day sunset window, and paid tiers are the only ones who will notice because free accounts were moved off the o-series long ago. Here's the part most of the coverage got half right. The retirement is a ChatGPT product change, so nothing you call over the API broke today. But the API is unaffected only until 11 December, which is the shutdown date sitting on OpenAI's own deprecations page for the o3 and o3-pro snapshots. Two clocks, two dates.
The short answer
OpenAI retired o3 from ChatGPT on 26 August 2026, closing a published 90 day
sunset window. Only paid tiers had access, so free accounts see nothing. Old
o3 threads continue on a current GPT-5 model. The API is a separate matter:
a deprecation entry filed 11 June 2026 shuts down o3-2025-04-16
and o3-pro-2025-06-10 on 11 December 2026, with
gpt-5.6-sol named as the replacement and
reasoning.mode: pro standing in for o3-pro.
What actually changed today
Nothing dramatic, if you’re a ChatGPT user. o3 disappeared from the model settings for Plus, Pro and Team. That’s it. Free and Go accounts have been on the 5.x line for months, so for most of the userbase this is invisible.
It’s the end of a queue rather than a surprise. GPT-4.5 went first, out of ChatGPT on 27 June 2026 after a 30 day window. o3 got 90 days, and today is day ninety. OpenAI’s framing across both was the same: pull older models with limited usage so the compute goes to newer ones.
Old conversations aren’t lost. OpenAI has continued threads on the corresponding current model through every swap this year, and the o3 retirement follows that pattern. Worth knowing that it never publishes which variant a retired thread inherits, and the reasoning behaviour of a GPT-5.6 tier is not the reasoning behaviour of o3. If a long o3 thread is load-bearing for you, export it. Takes a minute.
The API is not unaffected. It’s just on a later date.
This is the bit worth your attention, because a lot of the write-ups today say the API is untouched and then stop. One of the pieces we read while checking this states plainly that o3 remains available through the API with no announced end date.
That isn’t right. OpenAI’s deprecations page lists an entry announced 11 June 2026 with a shutdown date of 11 December 2026, covering o3-2025-04-16 and o3-pro-2025-06-10. Recommended replacement for both: gpt-5.6-sol, and for the pro snapshot, gpt-5.6-sol with reasoning.mode set to pro.
So the accurate version is that today’s change doesn’t touch your API traffic, and you have roughly fifteen weeks before something does. That’s a genuinely generous runway by 2026 standards, 183 days from notice to shutdown against 90 for the chat product. Use it.
Migrating off o3, honestly
The mechanical part is small. Swap the model string, and if you were calling o3-pro, add the reasoning mode parameter. It isn’t a family change, and the pinned snapshot names mean you know exactly which strings to grep for in your codebase.
The part that will actually cost you time is evaluation. o3 has a particular shape to its reasoning that plenty of teams built prompts around, and a replacement OpenAI calls equivalent is a recommendation, not a guarantee your eval suite holds. Budget for a rerun, not a find-and-replace.
One trap specific to this migration. If you prototype the new prompts inside ChatGPT and then ship them to the API, you may not be testing what you deploy. OpenAI retuned GPT-5.6 Sol for ChatGPT chat only on 6 August and said in the announcement that this version lives only in the Chat experience. Same name, different tuning. We wrote about that split when it happened and it applies squarely here: your o3 replacement in the chat window is not the gpt-5.6-sol your code calls. Test in the API.
What we could not confirm
Two things, and we’d rather flag them than paper over them.
We could not read OpenAI’s ChatGPT release notes directly. The help centre returned a 403 to us, so the ChatGPT-side dates and the thread continuation behaviour here come from the coverage that quoted those notes, plus the arithmetic of the published sunset windows. The API dates are a different story. Those come off the developer deprecations page and we read that one ourselves.
The announcement date for the ChatGPT retirement is muddier than it should be. Some outlets place the notice at the end of May, which is what the 30 and 90 day windows work out to, and others date their coverage to 3 June. We’ve used the sunset window lengths and the hard retirement dates, since those are consistent everywhere and they’re the numbers you’d actually plan against.
If you’re keeping a list of what OpenAI has pulled this year, the o-series leaving ChatGPT is the tidier end of it. The GPT-5.6 rollout with Sol, Terra and Luna is where the picker went instead, and Microsoft ran its own retirement wave on the Copilot side a week ago. Everybody is consolidating. December is the date to put in the calendar.
Sources
The 26 August 2026 ChatGPT retirement date for o3, the 90 day sunset window, the 27 June 2026 date and 30 day window for GPT-4.5, and the paid-tier-only access are reported by Android Authority, gHacks and The GPU Trade, all citing OpenAI’s ChatGPT release notes. The API shutdown date of 11 December 2026 for the o3 and o3-pro snapshots, the 11 June 2026 announcement date and the gpt-5.6-sol replacement recommendations come from OpenAI’s own deprecations page. The claim that o3 has no announced API end date, which the deprecations page contradicts, appears in OrcaRouter’s write-up, which is also where the thread export advice comes from.
Frequently asked questions
When was OpenAI o3 retired from ChatGPT?
On 26 August 2026. OpenAI announced the retirement with a 90 day sunset window, and the same notice gave GPT-4.5 a 30 day window that expired on 27 June 2026. Both models were only reachable by paid ChatGPT tiers by then, so free and Go accounts saw no change at all.
Does the ChatGPT retirement break my API calls to o3?
No. The 26 August change is a ChatGPT product change and it does not touch the API. What does touch the API is a separate deprecation entry announced on 11 June 2026, which sets 11 December 2026 as the shutdown date for the o3-2025-04-16 and o3-pro-2025-06-10 snapshots. Until that date your calls keep returning normally. After it they will not.
What does OpenAI recommend instead of o3 in the API?
The deprecations page names gpt-5.6-sol as the replacement for o3-2025-04-16. For o3-pro-2025-06-10 it names gpt-5.6-sol again, this time with reasoning.mode set to pro. So the migration is a model string change plus, for o3-pro traffic, one extra parameter rather than a move to a different family.
What happens to my existing ChatGPT conversations that used o3?
They keep working. OpenAI has handled every recent swap the same way, continuing old threads on the corresponding current GPT-5 model rather than freezing or deleting them. What it has not published is which specific GPT-5 variant an o3 thread lands on, so if a long o3 conversation matters to you, export it rather than assume the reasoning depth carries over unchanged.
Is the ChatGPT version of gpt-5.6-sol the same thing as the API one?
No, and this catches people migrating off o3. On 6 August 2026 OpenAI shipped a version of GPT-5.6 Sol tuned specifically for everyday ChatGPT conversation, and said in the announcement that it would only be available in the Chat experience. If you tune a prompt in the chat window and paste it into an API call to gpt-5.6-sol, you are not talking to the same tuning.